OPTIMIZATION !

Progs · Post by **Progs** » Thu Oct 14, 2004 9:36 am

I wish to see better assembler output....
I suppose that now ther is no optimization at all... only Line by line translation to asm.

Here PureBasic generated code

Code: Select all

; *mae_selectedlinkedlist\count = *mae_selectedlinkedlist\count - 1 
MOV    ebp,dword [p_mae_selectedlinkedlist]
MOV    ebx,dword [ebp+20]
ADD    ebx,-1
PUSH   ebx
MOV    ebp,dword [p_mae_selectedlinkedlist]
POP    eax
MOV    dword [ebp+20],eax

i think this code is bad....
it should be like

Code: Select all

MOV ebp,dword [p_mae_selectedlinkedlist]
MOV eax,dword [ebp + 20]
DEC eax
MOV dword [ebp + 20], eax

or even like

Code: Select all

MOV ebp,dword [p_mae_selectedlinkedlist]
DEC dword [ebp + 20]

sorry if im wrong...

Blade · Post by **Blade** » Thu Oct 14, 2004 12:56 pm

I wonder if a thirty party asm-optimized can be added to the compiling pipeline...
Or can already be done manually?

Progs · Post by **Progs** » Thu Oct 14, 2004 1:14 pm

You can write directly in asm

Example... pure function

Code: Select all

Procedure.l mae_linkedlist_next()
    If *mae_selectedlinkedlist\nxt = 0 
        *mae_selectedlinkedlist\pointer = 0
        ProcedureReturn 0
    EndIf 
    
    *mae_selectedlinkedlist\pointer = *mae_selectedlinkedlist\nxt
    *mae_selectedlinkedlist\nxt = *mae_selectedlinkedlist\pointer\nxt

    ProcedureReturn (*mae_selectedlinkedlist\pointer-*mae_selectedlinkedlist\datasize)
EndProcedure

My asm version

Code: Select all

Procedure.l mae_linkedlist_next() 
    !MOV ebp,dword [p_mae_selectedlinkedlist] 
    !MOV ebx,dword [ebp+12]
    !CMP ebx, 0
    !JNE _my_ok 
    !MOV dword [ebp+8],0
    !XOR eax,eax
    !JMP _my_end 
    !_my_ok: 
    ;*mae_selectedlinkedlist\pointer = *mae_selectedlinkedlist\nxt 
    !MOV eax,dword [ebp+12]
    !MOV dword [ebp+8],eax 
    ;*mae_selectedlinkedlist\nxt = *mae_selectedlinkedlist\pointer\nxt   
    !MOV ebp,dword [ebp+8] 
    !MOV eax,dword [ebp] ;eax = *nxt 
    !MOV ebp,dword [p_mae_selectedlinkedlist];ebx 
    !MOV dword [ebp+12], eax 
    ;return
    !MOV eax,dword [ebp+8] 
    !NOP
    !SUB eax,dword [ebp+16]
    ProcedureReturn
    !_my_end:
EndProcedure

ASM version is about 50% faster !

fsw · Post by **fsw** » Thu Oct 14, 2004 5:16 pm

Well, PureBasic's website states:

The key features of PureBasic are portability (Windows, AmigaOS and Linux are currently fully supported), the production of very fast and highly optimized executables and, of course, the very simple BASIC syntax.

I'm not a asm pro (so are most of pb users...), so I can't say if this is true with every bit of code pb produces.
But if you find some stuff just post it... it's appreciated.
Good hunting

Num3 · Post by **Num3** » Fri Oct 15, 2004 9:42 am

On Amiga there was a small program that would optimize assembler source code.

You would just select the source, and it would parse it and compare it against a built in optimization database, and would replace has much code has possible.

Moonshine · Post by **Moonshine** » Fri Oct 15, 2004 11:01 am

Is there anything that wll do the same for x86 on a PC though?

freedimension · Post by **freedimension** » Fri Oct 15, 2004 12:58 pm

Moonshine wrote:Is there anything that wll do the same for x86 on a PC though?

Num3 · Post by **Num3** » Fri Oct 15, 2004 1:42 pm

Oh, that is an evil one!

Post by **Dare2** » Fri Oct 15, 2004 1:45 pm

lol. Cruel.

There is always a downside to a post like that - your own questions could end up get links back to that post!

freedimension · Post by **freedimension** » Fri Oct 15, 2004 1:55 pm

Dare2 wrote:lol. Cruel.

There is always a downside to a post like that - your own questions could end up get links back to that post!

I never ask questions, do I????

benny · Post by **benny** » Fri Oct 15, 2004 3:30 pm

@freedimension:

At least from now on, you've laid a lot of pressure on yourself before you
can ask a question